The problem of subsalt imaging in the presence of salt-boundary uncertainty is considered. In this case, the salt-body model should be considered by a probability density function rather than by a unique deterministic function. We discuss a way to look at seismic imaging using the path-integral concept. The method computes the image by summing the contributions of individual images computed for different statistical realizations of the salt-body boundaries. It samples different realizations instead of relying on only one model derived from manual or automatic picking. The focusing mechanism is achieved by a weighting function (probability amplitude), which is designed to emphasize contributions from models close to the stationary one and to suppress contributions from unlikely models. The presented examples demonstrate principles and feasibility of the new concept.